Message ID | 20230614072548.996940-1-linus.walleij@linaro.org (mailing list archive) |
---|---|
State | New |
Headers | show |
Series | [v5] Documentation/mm: Initial page table documentation | expand |
Linus Walleij <linus.walleij@linaro.org> writes: > This is based on an earlier blog post at people.kernel.org, > it describes the concepts about page tables that were hardest > for me to grasp when dealing with them for the first time, > such as the prevalent three-letter acronyms pfn, pgd, p4d, > pud, pmd and pte. > > I don't know if this is what people want, but it's what I would > have wanted. The wording, introduction, choice of initial subjects > and choice of style is mine. > > I discussed at one point with Mike Rapoport to bring this into > the kernel documentation, so here is a small proposal. > > The current form is augmented in response to feedback from > Mike Rapoport, Matthew Wilcox, Jonathan Cameron, Kuan-Ying Lee, > Randy Dunlap and Bagas Sanjaya. > > Cc: Matthew Wilcox <willy@infradead.org> > Cc: Randy Dunlap <rdunlap@infradead.org> > Reviewed-by: Mike Rapoport <rppt@kernel.org> > Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com> > Reviewed-by: Bagas Sanjaya <bagasdotme@gmail.com> > Link: https://people.kernel.org/linusw/arm32-page-tables > Signed-off-by: Linus Walleij <linus.walleij@linaro.org> So I think this has gone around enough and have decided to pick it up. If there are parts that people aren't happy with, we can surely fix them up as we go on. Meanwhile, it's good to see an effort to fill in one of the gaps here, thanks for doing it. Thanks, jon
On mercoledì 14 giugno 2023 09:25:48 CEST Linus Walleij wrote: > This is based on an earlier blog post at people.kernel.org, > it describes the concepts about page tables that were hardest > for me to grasp when dealing with them for the first time, > such as the prevalent three-letter acronyms pfn, pgd, p4d, > pud, pmd and pte. > I don't know if this is what people want, but it's what I would > have wanted. The wording, introduction, choice of initial subjects > and choice of style is mine. > > I discussed at one point with Mike Rapoport to bring this into > the kernel documentation, so here is a small proposal. > > The current form is augmented in response to feedback from > Mike Rapoport, Matthew Wilcox, Jonathan Cameron, Kuan-Ying Lee, > Randy Dunlap and Bagas Sanjaya. > > Cc: Matthew Wilcox <willy@infradead.org> > Cc: Randy Dunlap <rdunlap@infradead.org> > Reviewed-by: Mike Rapoport <rppt@kernel.org> > Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com> > Reviewed-by: Bagas Sanjaya <bagasdotme@gmail.com> > Link: https://people.kernel.org/linusw/arm32-page-tables > Signed-off-by: Linus Walleij <linus.walleij@linaro.org> > --- I am writing to express my dissent regarding the proposal to add basic information about the role of hierarchical (multi-level) page tables in mapping virtual memory to physical page frames. While I understand the importance of documentation, I believe that including such fundamental operating system concepts in the specialized Linux kernel documentation would be redundant and unnecessary. The proposed addition appears to be a combination of trivia and a basic Operating Systems I course that one might encounter during their second year as an undergraduate student studying Computer Science or Computer Engineering. AFAIK, these concepts are already taught extensively to individuals pursuing a B.Sc. degree in Computer Science or a related field, both in Italy, where I live, and elsewhere. Therefore, it seems unlikely that Linux kernel developers would be unfamiliar with such fundamental topics, such as the mapping of virtual memory to physical page frames using multi-level (hierarchical) page tables. I question the target audience of this documentation. How can we expect any developer working with Linux to be unaware of such basic concepts? Adding documentation about these foundational concepts would create a precedent, potentially leading to further documentation on other fundamental abstractions like "task," "multi-threading," and "scheduling" – concepts that are integral to kernel management. The inclusion of such basic topics could quickly clutter up the specialized Linux kernel documentation. Let us not forget that there is a wealth of resources available outside the Linux kernel documentation. Books on OS theory or online courses from esteemed universities can easily provide individuals with the necessary knowledge on these fundamental concepts. Encouraging developers to explore these external resources fosters a culture of continuous learning and self-improvement, benefiting the entire Linux development community. In conclusion, I respectfully oppose the proposal to add basic operating system concepts, such as the hierarchical page tables, to the official Linux kernel documentation. I believe that such information is readily accessible through existing resources and that the specialized documentation should focus on advanced topics and unique aspects specific to the Linux kernel. Thank you for considering my perspective. Regards, Fabio M. De Francesco P.S.: The only parts I find enough interesting are those regarding the names of the types and other few bits of information, only because these are indeed Linux kernel focused and may not be found in the above-mentioned wealth of resources available outside, like Tanenbaum's and Silberschatz's books (however I'm not entirely sure they miss those information). > ChangeLog v4->v5: > - Drop the word "target" from the paragraph about virtual > addresses as pointed out by Matthew Wilcox. > - Drop "program counter" mention in paragraph about physical and > virtual addresses as pointed out by Matthew Wilcox. > - Update the changelog below to reflect who provided which > feedback so everybode can see that their feedback is being > taken into account. > - Collect Mike Rapoports Review tag. > ChangeLog v3->v4: > - Singularis to pluralis fix pointed out by Jonathan Cameron > - Reword the origin story about hierarchical page tables a bit > inspired by the input from Mike Rapoport. > ChangeLog v2->v3: > - Fix the page size example, also have examples for both 4K and > 16K pages since people will confront these in response to > feedback from Kuan-Ying Lee. > - Add a section explaining a bit why we have hierarchical > page tables at all. > ChangeLog v1->v2: > - Fixed speling mistakes > - Copyedit the paragraph on page frame numbers in response > to feedback from Matthew Wilcox. > - Reverse the arrows in the page table hierarchy illustration in > response to feedback from Matthew Wilcox. > - Reverse the order of description of the page hierarchy levels in > response to feedback from Matthew Wilcox. > - Create a new section for folding > - Emphasize that architectures should try to be page hierarchy > neutral in response to feedback from Mike Rapoport. > - Trying to better describe the fact that the lowest page table PTE > is called like that for historical reasons, in response to > sevaral comments on earlier blog posts on the subject. > --- > Documentation/mm/page_tables.rst | 149 +++++++++++++++++++++++++++++++ > 1 file changed, 149 insertions(+) > > diff --git a/Documentation/mm/page_tables.rst > b/Documentation/mm/page_tables.rst index 96939571d7bc..7840c1891751 100644 > --- a/Documentation/mm/page_tables.rst > +++ b/Documentation/mm/page_tables.rst > @@ -3,3 +3,152 @@ > =========== > Page Tables > =========== > + > +Paged virtual memory was invented along with virtual memory as a concept in > +1962 on the Ferranti Atlas Computer which was the first computer with paged > +virtual memory. The feature migrated to newer computers and became a de facto > +feature of all Unix-like systems as time went by. In 1985 the feature was > +included in the Intel 80386, which was the CPU Linux 1.0 was developed on. + > +Page tables map virtual addresses as seen by the CPU into physical addresses > +as seen on the external memory bus. > + > +Linux defines page tables as a hierarchy which is currently five levels in > +height. The architecture code for each supported architecture will then > +map this to the restrictions of the hardware. > + > +The physical address corresponding to the virtual address is often referenced > +by the underlying physical page frame. The **page frame number** or **pfn** > +is the physical address of the page (as seen on the external memory bus) > +divided by `PAGE_SIZE`. > + > +Physical memory address 0 will be *pfn 0* and the highest pfn will be > +the last page of physical memory the external address bus of the CPU can > +address. > + > +With a page granularity of 4KB and a address range of 32 bits, pfn 0 is at > +address 0x00000000, pfn 1 is at address 0x00001000, pfn 2 is at 0x00002000 > +and so on until we reach pfn 0xfffff at 0xfffff000. With 16KB pages pfs are > +at 0x00004000, 0x00008000 ... 0xffffc000 and pfn goes from 0 to 0x3fffff. > + > +As you can see, with 4KB pages the page base address uses bits 12-31 of the > +address, and this is why `PAGE_SHIFT` in this case is defined as 12 and > +`PAGE_SIZE` is usually defined in terms of the page shift as `(1 << > PAGE_SHIFT)` + > +Over time a deeper hierarchy has been developed in response to increasing > memory +sizes. When Linux was created, 4KB pages and a single page table > called +`swapper_pg_dir` with 1024 entries was used, covering 4MB which > coincided with +the fact that Torvald's first computer had 4MB of physical > memory. Entries in +this single table were referred to as *PTE*:s - page > table entries. + > +The software page table hierarchy reflects the fact that page table hardware > has +become hierarchical and that in turn is done to save page table memory > and +speed up mapping. > + > +One could of course imagine a single, linear page table with enormous amounts > +of entries, breaking down the whole memory into single pages. Such a page > table +would be very sparse, because large portions of the virtual memory > usually +remains unused. By using hierarchical page tables large holes in the > virtual +address space does not waste valuable page table memory, because it > will suffice +to mark large areas as unmapped at a higher level in the page > table hierarchy. + > +Additionally, on modern CPUs, a higher level page table entry can point > directly +to a physical memory range, which allows mapping a contiguous range > of several +megabytes or even gigabytes in a single high-level page table > entry, taking +shortcuts in mapping virtual memory to physical memory: there > is no need to +traverse deeper in the hierarchy when you find a large mapped > range like this. + > +The page table hierarchy has now developed into this:: > + > + +-----+ > + | PGD | > + +-----+ > + | > + | +-----+ > + +-->| P4D | > + +-----+ > + | > + | +-----+ > + +-->| PUD | > + +-----+ > + | > + | +-----+ > + +-->| PMD | > + +-----+ > + | > + | +-----+ > + +-->| PTE | > + +-----+ > + > + > +Symbols on the different levels of the page table hierarchy have the > following +meaning beginning from the bottom: > + > +- **pte**, `pte_t`, `pteval_t` = **Page Table Entry** - mentioned earlier. > + The *pte* is an array of `PTRS_PER_PTE` elements of the `pteval_t` type, > each + mapping a single page of virtual memory to a single page of physical > memory. + The architecture defines the size and contents of `pteval_t`. > + > + A typical example is that the `pteval_t` is a 32- or 64-bit value with the > + upper bits being a **pfn** (page frame number), and the lower bits being > some + architecture-specific bits such as memory protection. > + > + The **entry** part of the name is a bit confusing because while in Linux > 1.0 + this did refer to a single page table entry in the single top level > page + table, it was retrofitted to be an array of mapping elements when > two-level + page tables were first introduced, so the *pte* is the lowermost > page + *table*, not a page table *entry*. > + > +- **pmd**, `pmd_t`, `pmdval_t` = **Page Middle Directory**, the hierarchy > right + above the *pte*, with `PTRS_PER_PMD` references to the *pte*:s. > + > +- **pud**, `pud_t`, `pudval_t` = **Page Upper Directory** was introduced > after + the other levels to handle 4-level page tables. It is potentially > unused, + or *folded* as we will discuss later. > + > +- **p4d**, `p4d_t`, `p4dval_t` = **Page Level 4 Directory** was introduced to > + handle 5-level page tables after the *pud* was introduced. Now it was > clear + that we needed to replace *pgd*, *pmd*, *pud* etc with a figure > indicating the + directory level and that we cannot go on with ad hoc names > any more. This + is only used on systems which actually have 5 levels of > page tables, otherwise + it is folded. > + > +- **pgd**, `pgd_t`, `pgdval_t` = **Page Global Directory** - the Linux kernel > + main page table handling the PGD for the kernel memory is still found in + > `swapper_pg_dir`, but each userspace process in the system also has its own > + memory context and thus its own *pgd*, found in `struct mm_struct` which + > in turn is referenced to in each `struct task_struct`. So tasks have memory > + context in the form of a `struct mm_struct` and this in turn has a + > `struct pgt_t *pgd` pointer to the corresponding page global directory. + > +To repeat: each level in the page table hierarchy is a *array of pointers*, > so +the **pgd** contains `PTRS_PER_PGD` pointers to the next level below, > **p4d** +contains `PTRS_PER_P4D` pointers to **pud** items and so on. The > number of +pointers on each level is architecture-defined.:: > + > + PMD > + --> +-----+ PTE > + | ptr |-------> +-----+ > + | ptr |- | ptr |-------> PAGE > + | ptr | \ | ptr | > + | ptr | \ ... > + | ... | \ > + | ptr | \ PTE > + +-----+ +----> +-----+ > + | ptr |-------> PAGE > + | ptr | > + ... > + > + > +Page Table Folding > +================== > + > +If the architecture does not use all the page table levels, they can be > *folded* +which means skipped, and all operations performed on page tables > will be +compile-time augmented to just skip a level when accessing the next > lower +level. > + > +Page table handling code that wishes to be architecture-neutral, such as the > +virtual memory manager, will need to be written so that it traverses all of > the +currently five levels. This style should also be preferred for > +architecture-specific code, so as to be robust to future changes. > -- > 2.40.1
"Fabio M. De Francesco" <fmdefrancesco@gmail.com> writes: > I question the target audience of this documentation. How can we expect any > developer working with Linux to be unaware of such basic concepts? Adding > documentation about these foundational concepts would create a precedent, > potentially leading to further documentation on other fundamental abstractions > like "task," "multi-threading," and "scheduling" – concepts that are integral > to kernel management. The inclusion of such basic topics could quickly clutter > up the specialized Linux kernel documentation. Someday, if we find ourselves in the position of having too much documentation, we can entertain patches to clean out material that is deemed to be too elementary for kernel developers. Before then, though, if we are worried about clutter, we may want to put more effort into addressing the large amount of duplicated and obsolete documentation in the kernel now. Until then, I see no reason to oppose the addition of material that, even if you don't personally find it helpful, may indeed be helpful to developers trying to come up to speed on just what the kernel is doing. Thanks, jon
On Sun, Jun 18, 2023 at 3:16 PM Fabio M. De Francesco <fmdefrancesco@gmail.com> wrote: > I am writing to express my dissent regarding the proposal to add basic > information about the role of hierarchical (multi-level) page tables in > mapping virtual memory to physical page frames. I have understood that some think this, perhaps the intro could use some dieting, what about sending a patch to make it look like you want it to? > The proposed addition appears to be a combination of trivia and a basic > Operating Systems I course that one might encounter during their second year > as an undergraduate student studying Computer Science or Computer Engineering. > > AFAIK, these concepts are already taught extensively to individuals pursuing a > B.Sc. degree in Computer Science or a related field, both in Italy, where I > live, and elsewhere. Knowing the audience is always the hard part of wording technical documentation, not the contents per se. I might fail, I might be slight off, my co-developers are there to help. Assuming that newcomers to the Linux kernel have formal academic background or specifically operating system education is a bit thick IMO, suffice to read pages 108-111 of Glyn Moody's book "Rebel Code" about the background of the network maintainer. There are a whole bunch of random people attracted to Linux development. Memory management may be different though? Mel having written his PhD thesis about the Linux VMM and all might set the bar higher for contributors. I don't know really. But the documentation is not there just for the MM contributors, as the MM primitives are found sprinkled all over the kernel. Yours, Linus Walleij
On lunedì 19 giugno 2023 10:16:56 CEST Linus Walleij wrote: > On Sun, Jun 18, 2023 at 3:16 PM Fabio M. De Francesco > > <fmdefrancesco@gmail.com> wrote: > > I am writing to express my dissent regarding the proposal to add basic > > information about the role of hierarchical (multi-level) page tables in > > mapping virtual memory to physical page frames. [...] > Assuming that newcomers to the Linux kernel have formal academic > background or specifically operating system education is a bit thick > IMO, suffice to read pages 108-111 of Glyn Moody's book > "Rebel Code" about the background of the network maintainer. > There are a whole bunch of random people attracted to Linux > development. Linus, I must admit that I have had a change of heart regarding the necessity of this documentation. This change came about after reading Jon's reply, as well as your own. However, it wasn't just because of the two of you. It was mainly due to my conversations with some colleagues I work with, who hold M.Sc. degrees in Computer Science. Despite not having a formal background in CS or CE myself, I have taken the time to self-teach the subject matter, which I expected them to be well-versed in. To my surprise, they only have a vague understanding of page tables and the fact that processes use addresses that may not correspond to physical locations. That's about it! Hence, I now fully support your initiative and want to express my gratitude for undertaking this task. The only thing I would prefer not to see is the historical reference to the first implementation of hierarchical page tables. After all, many concepts implemented in Linux are derived or adapted from existing knowledge or implementations in other kernels. However, I can also understand why you prefer to have it as an introduction to the subject. Once again thanks, Fabio
Hi Fabio! thanks for your reply! The ways of technical documentation are never easy, but what we are using right now is the socratic method, dialogue at its best, which is pretty much the best way I know. WRT the problem of education: In gloomy days I have been referring to something I tongue-in-cheek call "the second software crisis" (not any established term) so in contrast with the first software crisis which was about the complexity of software development outgrowing hardware development, the second software crisis is due to software developers losing contact and knowledge of hardware, with big white spots on their mental map so that is part of what I am trying to fix here. Best regards, Linus Walleij
diff --git a/Documentation/mm/page_tables.rst b/Documentation/mm/page_tables.rst index 96939571d7bc..7840c1891751 100644 --- a/Documentation/mm/page_tables.rst +++ b/Documentation/mm/page_tables.rst @@ -3,3 +3,152 @@ =========== Page Tables =========== + +Paged virtual memory was invented along with virtual memory as a concept in +1962 on the Ferranti Atlas Computer which was the first computer with paged +virtual memory. The feature migrated to newer computers and became a de facto +feature of all Unix-like systems as time went by. In 1985 the feature was +included in the Intel 80386, which was the CPU Linux 1.0 was developed on. + +Page tables map virtual addresses as seen by the CPU into physical addresses +as seen on the external memory bus. + +Linux defines page tables as a hierarchy which is currently five levels in +height. The architecture code for each supported architecture will then +map this to the restrictions of the hardware. + +The physical address corresponding to the virtual address is often referenced +by the underlying physical page frame. The **page frame number** or **pfn** +is the physical address of the page (as seen on the external memory bus) +divided by `PAGE_SIZE`. + +Physical memory address 0 will be *pfn 0* and the highest pfn will be +the last page of physical memory the external address bus of the CPU can +address. + +With a page granularity of 4KB and a address range of 32 bits, pfn 0 is at +address 0x00000000, pfn 1 is at address 0x00001000, pfn 2 is at 0x00002000 +and so on until we reach pfn 0xfffff at 0xfffff000. With 16KB pages pfs are +at 0x00004000, 0x00008000 ... 0xffffc000 and pfn goes from 0 to 0x3fffff. + +As you can see, with 4KB pages the page base address uses bits 12-31 of the +address, and this is why `PAGE_SHIFT` in this case is defined as 12 and +`PAGE_SIZE` is usually defined in terms of the page shift as `(1 << PAGE_SHIFT)` + +Over time a deeper hierarchy has been developed in response to increasing memory +sizes. When Linux was created, 4KB pages and a single page table called +`swapper_pg_dir` with 1024 entries was used, covering 4MB which coincided with +the fact that Torvald's first computer had 4MB of physical memory. Entries in +this single table were referred to as *PTE*:s - page table entries. + +The software page table hierarchy reflects the fact that page table hardware has +become hierarchical and that in turn is done to save page table memory and +speed up mapping. + +One could of course imagine a single, linear page table with enormous amounts +of entries, breaking down the whole memory into single pages. Such a page table +would be very sparse, because large portions of the virtual memory usually +remains unused. By using hierarchical page tables large holes in the virtual +address space does not waste valuable page table memory, because it will suffice +to mark large areas as unmapped at a higher level in the page table hierarchy. + +Additionally, on modern CPUs, a higher level page table entry can point directly +to a physical memory range, which allows mapping a contiguous range of several +megabytes or even gigabytes in a single high-level page table entry, taking +shortcuts in mapping virtual memory to physical memory: there is no need to +traverse deeper in the hierarchy when you find a large mapped range like this. + +The page table hierarchy has now developed into this:: + + +-----+ + | PGD | + +-----+ + | + | +-----+ + +-->| P4D | + +-----+ + | + | +-----+ + +-->| PUD | + +-----+ + | + | +-----+ + +-->| PMD | + +-----+ + | + | +-----+ + +-->| PTE | + +-----+ + + +Symbols on the different levels of the page table hierarchy have the following +meaning beginning from the bottom: + +- **pte**, `pte_t`, `pteval_t` = **Page Table Entry** - mentioned earlier. + The *pte* is an array of `PTRS_PER_PTE` elements of the `pteval_t` type, each + mapping a single page of virtual memory to a single page of physical memory. + The architecture defines the size and contents of `pteval_t`. + + A typical example is that the `pteval_t` is a 32- or 64-bit value with the + upper bits being a **pfn** (page frame number), and the lower bits being some + architecture-specific bits such as memory protection. + + The **entry** part of the name is a bit confusing because while in Linux 1.0 + this did refer to a single page table entry in the single top level page + table, it was retrofitted to be an array of mapping elements when two-level + page tables were first introduced, so the *pte* is the lowermost page + *table*, not a page table *entry*. + +- **pmd**, `pmd_t`, `pmdval_t` = **Page Middle Directory**, the hierarchy right + above the *pte*, with `PTRS_PER_PMD` references to the *pte*:s. + +- **pud**, `pud_t`, `pudval_t` = **Page Upper Directory** was introduced after + the other levels to handle 4-level page tables. It is potentially unused, + or *folded* as we will discuss later. + +- **p4d**, `p4d_t`, `p4dval_t` = **Page Level 4 Directory** was introduced to + handle 5-level page tables after the *pud* was introduced. Now it was clear + that we needed to replace *pgd*, *pmd*, *pud* etc with a figure indicating the + directory level and that we cannot go on with ad hoc names any more. This + is only used on systems which actually have 5 levels of page tables, otherwise + it is folded. + +- **pgd**, `pgd_t`, `pgdval_t` = **Page Global Directory** - the Linux kernel + main page table handling the PGD for the kernel memory is still found in + `swapper_pg_dir`, but each userspace process in the system also has its own + memory context and thus its own *pgd*, found in `struct mm_struct` which + in turn is referenced to in each `struct task_struct`. So tasks have memory + context in the form of a `struct mm_struct` and this in turn has a + `struct pgt_t *pgd` pointer to the corresponding page global directory. + +To repeat: each level in the page table hierarchy is a *array of pointers*, so +the **pgd** contains `PTRS_PER_PGD` pointers to the next level below, **p4d** +contains `PTRS_PER_P4D` pointers to **pud** items and so on. The number of +pointers on each level is architecture-defined.:: + + PMD + --> +-----+ PTE + | ptr |-------> +-----+ + | ptr |- | ptr |-------> PAGE + | ptr | \ | ptr | + | ptr | \ ... + | ... | \ + | ptr | \ PTE + +-----+ +----> +-----+ + | ptr |-------> PAGE + | ptr | + ... + + +Page Table Folding +================== + +If the architecture does not use all the page table levels, they can be *folded* +which means skipped, and all operations performed on page tables will be +compile-time augmented to just skip a level when accessing the next lower +level. + +Page table handling code that wishes to be architecture-neutral, such as the +virtual memory manager, will need to be written so that it traverses all of the +currently five levels. This style should also be preferred for +architecture-specific code, so as to be robust to future changes.